Abstract

The messages exchanged in computational systems fall into two broad categories: data and control messages. Data messages transport data exchanged between tasks at the application layer. In computational systems, several other messages do not belong directly to the application layer. Those messages called control messages are related to the system management. The state-of-the-art shows a grow in the number of proposals employing dedicated NoC architectures for control messages. Dedicated NoCs are employed for, e.g., cache management, security, resource allocation, fault-tolerance, circuit switching, and system monitoring. A common limitation of these proposals is specialization. For example, it is not possible to use a dedicated NoC designed for cache coherence protocols to fault tolerance purposes. This paper presents the BrNoC, which can be employed for different specialized control messages, supporting different protocols. The flexibility of the proposal comes from a dedicated NoC having broadcast as the default transmission mode. The paper presents the BrNoC used in three different scenarios: fault-tolerance, security, management. Results evaluate each scenario, considering the performance, latency, and area cost. The results justify the adoption of the BrNoC as a dedicated NoC for control messages in many-core systems.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call